Bake Blue Box into AI coding agents to detect issues early
Matt Gibiec recommends baking Blue Box into AI coding agents so issues are detected early and developers become productive.
42,922 results · Most relevant
Matt Gibiec recommends baking Blue Box into AI coding agents so issues are detected early and developers become productive.
Matt Gibiec advises that when fixing a bug, you should verify the change doesn't cause issues elsewhere in your environment before pushing it, and if a fix breaks one service, check the dependency map of all other services before deploying.
Matt Gibiec advises that when suggesting a fix, you should use the context of your environment and monitored applications so the fix addresses what's actually happening, rather than only fixing the particular issue breaking the agent.
Abhi Arya recommends logging the prompts sent from the model in every MCP session, and asking team members (e.g., sales) to export their Claude Code sessions so you can understand how the model was behaving, where humans overrode the agent, where model confidence ran low, and where it asked questions.
Log the prompts sent from the model in every MCP session and review exported Claude Code sessions to find where humans overrode the agent, where model confidence ran low, and where it asked questions. Reducto even asked its sales team and other team members to export their Claude Code sessions.
Dr. Elissa Epel recommends using the mental frame 'be the lion instead of the gazelle' to shift into a challenge response during stress — reminding oneself to approach the stressor as the predator rather than the prey.
Tara Sesha recommends shipping products internally first and having everyone use them before external release, checking whether it retains well, whether people find something delightful about it, and whether it unlocks a new use case of the models.
Nan Yu recommends using a chief-of-staff agent that manages all the other agents rather than managing each one individually.
Check your real engineering metrics: median time to review a PR, time to reproduce user bug reports, time to debug code, who keeps docs current, and who maintains test suites. Pull your DORA metric — the time from code written to time in production likely dwarfs the time actually making engineering changes.
You can have animations and then tell Figma Make to be driven based off of your scroll, for example six sections each animating when they come into view.
The best animation export format for a personal business website depends on what you are animating: for simpler animations you can probably just export as CSS, while for more visual, heavier animations with lots of images and movement, MP4 might be better.
hedged with 'probably' and 'might'
When making an interaction, highlight the frames and bring them over from context so the transitions register; the presenter usually brings things in small chunks and sections rather than handing over a massive prototype.
The speaker recommends adding motion into components so that the animation is the same across all variants, increasing consistency throughout a toolkit.
Matt Owens says AI is most helpful for distillation, sense-making, summarization, and proposal writing. He creates AI skills that ingest specific client data to produce outcomes he can get into a presentation more quickly and iterate from there — he calls this infinitely helpful.
Use the Strands agent state (tool context) to persist tool information and the memory pointer ID inside the context window; create tools with the tool decorator, e.g., a fetch_application_logs tool with tool_context agent state.
Use the sliding window conversation management strategy to keep only the most recent messages; use the summarization conversation manager to summarize older messages while keeping recent ones; or combine summarization of older messages with retention of recent messages as a hybrid strategy. Configure the summarizing manager with a summary ratio (e.g., 50%) and preserve the last N messages (e.g., 4).
A second tool can analyze the stored logs and report back to the LLM a simple verdict (e.g., 'the logs are okay') without the LLM ever seeing the raw logs.
Use the Strands Agents framework (open source, free, model-agnostic, maintained by AWS) to build agent loops: a single line of code creates the entire agent loop, including nodes and graph, behind the scenes. To use a different model provider (OpenAI, Llama/Ollama, Gemini, Anthropic), you only need to import the provider library and add one line for the binding.
In his demo, Coach Kyle gets the woman's attention from a distance while she is still walking, smiles the whole way through, and comes to a stop together with her before starting the conversation.
For athletes in bent-position sports, strengthen athletes in highly bent trunk and hip flexion positions instead of changing their sport-specific running technique: use very intense trunk flexion exercises such as the drop and catch RDL, and use unilateral exercises as much as possible to replicate the lunge and sprint movement.
Choose sources → to get a library of your own and a daily brief.